Tweaktown : Nvidia Jetson (Tegra K1)

[coercitiv]

There is nothing to deduce. Techspot clearly stated that they did all their published measurements on AC power, and when they switched to battery power, performance went down by ~ 10% due to less aggressive boosting of CPU/GPU operating frequencies.

Oh but there is. If we read their exact words we see there's enough to interpret there, and you did just that.

When the tablet is benchmarked on battery power, I noticed a roughly 10% performance drop as the Mullins APU switches to using a more aggressive boost algorithm in attempt to lengthen battery life.

What you consider "higher clocks on AC power" might just be a more aggressive power scheme that delays boosting to higher clocks in order to lengthen battery life. We have no indication how this phenomenon will be observed in retail products. From the little we know for now, it may even be user configurable via Windows Power Options.

Anyway, enough about Mullins on the K1 thread. Xiaomi's MiPad looks like a nice win for Nvidia, I'm really curious to see benchmarks and battery life for TK1 in a tablet form factor.

Hopefully more OEMs will follow soon.

 

[monstercameron]

I had a zune hd with tegra 1 and just picked up a 2012 nexus 7, it's ok (brushing up on my tegra knowledge), want to pick up a shield or cheap tegra 4 tablet(can't find many).
The point of this anecdote, people like tegra but it becomes obsolete too fast(not saying that others dont also). Hopefully tk1v1/2 changes that.

 

[TrulyUncouth]

I had a zune hd with tegra 1 and just picked up a 2012 nexus 7, it's ok (brushing up on my tegra knowledge), want to pick up a shield or cheap tegra 4 tablet(can't find many).
The point of this anecdote, people like tegra but it becomes obsolete too fast(not saying that others dont also). Hopefully tk1v1/2 changes that.

I think the real benefit to Tegra going forward will be badass games like Portal and Half-life 2 coming to them. When I saw the announce for those two games I had to hold off on ordering a shield and even then I only held off because I really want the TK1 version. And this is coming from a guy who currently has an AMD desktop card and can't use the remote play feature.

If they deliver the power numbers they claim then NV will have a ridiculously good tablet/handheld chip. If they keep the performance up and get their modems in order next year then they could really give qualcomm a run for their money. Exciting times, assuming they deliver on power usage.

 

[xpea]

I think the real benefit to Tegra going forward will be badass games like Portal and Half-life 2 coming to them. When I saw the announce for those two games I had to hold off on ordering a shield and even then I only held off because I really want the TK1 version. And this is coming from a guy who currently has an AMD desktop card and can't use the remote play feature.

If they deliver the power numbers they claim then NV will have a ridiculously good tablet/handheld chip. If they keep the performance up and get their modems in order next year then they could really give qualcomm a run for their money. Exciting times, assuming they deliver on power usage.

Fully agree.
In phones, QC should keep the lead due do their strong modem tech/integration but in high-end tablets, I expect Nvidia taking a good chunk of market share. Even more if we take into account 64bit SoCs where QC faces r&d delay.
And I don't even talk about Erista (Denver + Maxwell on 20nm for mid 2015 devices), with another 100% jump in GPU performance, will be too much for all competitors...

 

[jdubs03]

Here are some more data points for the TK1v1:

http://www.evolife.cn/html/2014/77075.html

Not too shabby, though its interesting to see the 3Dmark score being ~4000 points lower than the MiPad.

 

[TrulyUncouth]

Here are some more data points for the TK1v1:

http://www.evolife.cn/html/2014/77075.html

Not too shabby, though its interesting to see the 3Dmark score being ~4000 points lower than the MiPad.

That is odd, especially since the Mipad is technically a lower binned version running at 2.2ghz, right?

Assuming qualcomm's 420 GPU ends up 40% faster than 330 like they say then this puts qualcomm and NV within easy range of each other I think. Hopefully that benchmark is simply wrong. But maybe that is the score it gets when run multiple times and can't maintain turbo speed? I can't wait for some real benches and hands on with these.

Especially once the shield version with K1 comes out.

EDIT: Actually, looking at that whole review the numbers seem too low for TK1 or too high for the 801. I wonder if the 16k 3d mark score quoted earlier was on different drivers? If 19k for 801 is accurate and 805 gets 40% faster then TK1's lead would evaporate- or am I reading wrong numbers?

 

[monstercameron]

Here are some more data points for the TK1v1:

http://www.evolife.cn/html/2014/77075.html

Not too shabby, though its interesting to see the 3Dmark score being ~4000 points lower than the MiPad.

waiting on a certain member to respond...

as some of us were stating it's too early to deduce anything but I will admit preliminary info is in tk1's favor...

 

[ams23]

Numbers on a finished product that is production-ready (ie. Xiaomi Mi Pad) can be trusted, there is nothing to deduce there. Numbers from an unknown source on an unknown test platform should be taken with a grain of salt. Note that most commercial devices will probably not use 2.3GHz as the maximum CPU clock operating frequency because the perf. per watt curve appears to be significantly better at closer to ~ 2.1-2.2GHz.

 

[blackened23]

Besides which, SOCs are highly configurable in terms of target performance and battery life. What that means is, the vendor can customize the SOC to have more of either.

That's why the iPad Air and Iphone 5s both use the same chip. But guess which one is faster. The iphone 5s doesn't perform as well as the full size ipad air in many benchmarks, because the A7 is configured to be faster in the ipad air since it is a larger device. This isn't a hard concept to understand - depending on the form factor being used, and how the vendor configures the chip, the CHIP WILL PERFORM DIFFERENTLY.

In a full size tablet, the Tegra K1 will perform a full potential. That means the Mi Pad is probably representative of how the K1 will perform in a tablet. If you put the same K1 in a smaller form factor such as a phone, then the vendor will obviously configure it differently and it may perform slightly less. I am surprised people are still unaware of this. There are many qualcomm 805 devices. Do they all perform the same? No. They perform differently depending on the form factor and how the vendor configures the SOC for the form factor.

 

[kpkp]

There aren't any Snapdragon 805 devices yet...
Most of the configurations you are talking about are done by the thermal limit/throttling. Would be interesting to see, how would a iPad A7 (with ipad configurations) perform in an iPhone, my guess... Not much different then the current iPhone version.

 

[blackened23]

It isn't even a throttling issue. Qualcomm 800 devices don't all perform the same, despite having the same SOC. The issue isn't throttling per se, the issue is taht vendors do and always have had the ability to customize SOCs per the form being used - if they want to tune it for a full tablet with more aggressive turbo, they can do that. If they want to tune it for more battery life or a phone size device, they can do that.

While the SOCs throttle, that isn't the issue in question. Vendors CAN customize SOCs for either battery life or performance. If you see a game console based on a qualcomm chip, it will obviously be tuned for performance. If you see a qualcomm 800 10 inch tablet, it will be tuned for performance. But if you see a qualcomm 800 phone, it will be tuned for higher battery life and slightly less performance.

So the point i'm getting at here, is that the K1 tested in the Mipad *IS* representative of what the silicon performs like in a tablet form factor. If the K1 is put in a shield type device, it will perform better than that. If the K1 is put in a smaller form factor such as a phone, it will perform slightly lower than that. The vendors will customize the SOC clockspeeds and turbo according to WHAT DEVICE it is put in. There are those questioning the performance of K1 tested by endgadget, for no real reason. Those numbers will be representative of K1 in a tablet, that's the point.

 

[kpkp]

You try so hard to dismiss the importance of thermal limits and then you make a comparison between a passive cooled device with an active cooled device...

Sure if your cooling is at a level where it does not limit the SoC performance, you can have more permissive governor (shield style). I hope you understand that I am not trying to say that they don't configure the governor, but that it doesn't have such a big impact as you are arguing. Just look at the Samsung cheating debacle from last year, yes, they did max-lock the turbo frequency and over-clocked the GPU, but all that wouldn't have that much of an impact if they wouldn't also ignore/raise the thermal limit. You can also look at the test of the S5 water vs. air, or the A7 tests done by Anand or even go way back and read the NExus 4 review... In most cases the CPU goes to max when needed and it doesn't go down until the thermal limit kicks in.

I would love to see a Snapdragon 800 with a max turbo of 1.8GHz or a A15 core at 1.2GHz (as ARM suggested for the earlier rev.) but that just doesn't happen. I guess marketing is to blame.

In general I have a feeling that you are giving most OEM to much credit, because mostly they optimize for benchmarks and less for battery. That's what leaks first and if the bench are slow, the nerds go crazy.

 

[altairauditore]

The Tegra K1 is 2x Apple A7. That's 99% confirmed.
Now, Adreno 330 also performs around the same (usually slower). So with Adreno 420, Qualcomm's 40% faster claim will be at its best (shader / computer performance is not everything).

Doesn't seem like TK1 will have a competitor, atleast in the tablet/console space.
Phones is another story. TK1 should manage it, but then it would probably be capped to 50-80% of it's performance.. for there to be really really good battery life.

 

[zebrax2]

http://www.cnet.com/news/nvidia-ceo-sees-future-in-cars-and-gaming-q-a/

It seems like NVidia is having a hard time penetrating the mobile market and instead focusing on other markets instead

 

[ams23]

http://www.cnet.com/news/nvidia-ceo-sees-future-in-cars-and-gaming-q-a/

It seems like NVidia is having a hard time penetrating the mobile market and instead focusing on other markets instead

Not exactly. NVIDIA is still very much focused on the consumer ultra mobile market (including tablets, high end smartphones, and gaming devices), but they are also focused on the automotive and embedded markets too. The only area that they are really not focusing on anymore is mainstream phones where price matters more than performance and feature set.

 

[caswow]

Not exactly. NVIDIA is still very much focused on the consumer ultra mobile market (including tablets, high end smartphones, and gaming devices), but they are also focused on the automotive and embedded markets too. The only area that they are really not focusing on anymore is mainstream phones where price matters more than performance and feature set.

oh yea. the good old "didnt want to be in there in the first place" trick.

 

[xpea]

oh yea. the good old "didnt want to be in there in the first place" trick.

Did you read the article ?
I think this cnet interview is very candid and honest. NV CEO even says T4i is a failure.

 

[caswow]

Did you read the article ?
I think this cnet interview is very candid and honest. NV CEO even says T4i is a failure.

I answered ams32 not what jhs said

 

[lefty2]

http://www.cnet.com/news/nvidia-ceo-sees-future-in-cars-and-gaming-q-a/

It seems like NVidia is having a hard time penetrating the mobile market and instead focusing on other markets instead

Yeah, I wonder if the Tegra division can ever be profitable if they can only compete in the automotive and high end tablet markets.

 

[Homeles]

I answered ams32 not what jhs said

And where exactly do you think ams32 got his information? Sheesh.

 

[ams23]

Yeah, I wonder if the Tegra division can ever be profitable if they can only compete in the automotive and high end tablet markets.

Tegra volumes will start to improve not just from the automotive space but also from the consumer space (tablets, tablet convertibles, high end smartphones, gaming devices, set-top boxes, etc.) and the embedded space (robotics, medical devices, military, etc). And now that the Tegra GPU architecture is in step with the modern-day GPU architecture, expenses can be much better controlled through leveraged investments. It is inevitable that Tegra will keep growing and will become profitable due to improved volumes in diversified and growing spaces and due to better control of expenses through leverage (although net profitability may be difficult to discern due to shared R&D expenses with other lines of business).

 

[lefty2]

Tegra volumes will start to improve not just from the automotive space but also from the consumer space (tablets, tablet convertibles, high end smartphones, gaming devices, set-top boxes, etc.) and the embedded space (robotics, medical devices, military, etc).

... then again maybe they won't. The only actual design wins we see are automotive and tablets (it's a well known fact that Nvidia smartphone design wins all involve contra revenue)

 

[ams23]

... then again maybe they won't. The only actual design wins we see are automotive and tablets (it's a well known fact that Nvidia smartphone design wins all involve contra revenue)

That makes no sense. Their smartphone design wins all involve "contra revenue"? LOL. Tegra 4i is a very low cost SoC design relatively speaking. The SoC die size is extremely small (~ 60 mm^2), even with built-in LTE modem. Tegra 4 also has a modest SoC die size (~ 80 mm^2, which is approximately 50% smaller than S800). And neither of these SoC's are fabbed on the latest-and-greatest HPM fab. process.

 

[lefty2]

That makes no sense. Their smartphone design wins all involve "contra revenue"? LOL. Tegra 4i is a very low cost SoC design relatively speaking. The SoC die size is extremely small (~ 60 mm^2), even with built-in LTE modem. Tegra 4 also has a modest SoC die size (~ 80 mm^2, which is approximately 50% smaller than S800). And neither of these SoC's are fabbed on the latest-and-greatest HPM fab. process.

That's irrelevant. The problem is there is no W-CDMA or TD-SCDMA support. Anyway Huang admits himself that Tegra 4i was a failure (first time I ever heard him admit defeat).

 

[ams23]

That's irrelevant. The problem is there is no W-CDMA or TD-SCDMA support.

Think about what you are saying. Tegra 4 in Xiaomi's Mi3 phone is used only with China Mobile. Tegra 4i in LG's G2 mini phone is used only outside of the USA. This has nothing to do with contra revenue. Prior generations of Tegra were always designed in a cost-conscious way.

Obviously Qualcomm and Mediatek will win the lion's share of smartphone business outside of Apple and Samsung because the former is able to cost-effectively bundle their modem with their SoC (and in fact some large cellular service providers such as Verizon and Sprint have legacy networks that require Qualcomm's WCDMA tech) and the latter is the lowest cost SoC supplier.

Until all the major cellular network providers migrate over to newer technologies such as VoLTE ( http://www.latimes.com/business/tec...-volte-what-exactly-is-it-20140522-story.html ), there will still be a need to rely on legacy 3G networks, and therefore Qualcomm's modem tech will still be required in many cases.